Daniel Pokorný

Posted on Jun 19 • Originally published at atomfoundry.dev

# The State of AI Recommendations Across Commerce 2026

#ai #webdev

Most businesses assume that if AI systems can understand their products, they will eventually recommend them.

Our latest research suggests the relationship may not be that simple.

Over the past several months, we analyzed 20,000 AI recommendations across five e-commerce categories:

Beauty
Supplements
Coffee
Pets
Home & Living

The goal was straightforward: To understand whether store readiness actually predicts recommendation behavior.

Or put differently: Do AI systems recommend the businesses that are best prepared for AI?

The answer surprised us.

The Assumption

Most discussions about AI visibility start with a reasonable belief.

If a business improves its structure, content, product information, trust signals, and machine readability, AI systems should be more likely to recommend it.

This assumption has fueled a growing industry around AI optimization, AI visibility, AI readiness, and AI commerce infrastructure.

But assumptions are not evidence.

We wanted to measure recommendation behavior directly.

What We Measured

For each category, we collected thousands of recommendations generated by AI systems in response to high-intent shopping questions.

We then compared recommendation frequency against each brand's AI Commerce Score™.

The expectation was simple: Higher scores should lead to more recommendations.

Instead, we found something very different.

Figure 1. AI Commerce Score™ and Recommendation Frequency™

Based on 20,000 AI recommendations across five ecommerce categories, recommendation frequency showed little to no meaningful relationship with AI Commerce Score™, suggesting that store readiness alone does not explain recommendation behavior.

The Surprising Result

Across all five categories, recommendation frequency showed little to no meaningful correlation with AI Commerce Score™.

Some highly recommended brands scored relatively poorly. Some highly optimized brands were rarely recommended.

The relationship was far weaker than expected.

This finding appeared repeatedly across Beauty, Supplements, Coffee, Pets, and Home & Living.

The implication is important.

Store readiness alone does not appear sufficient to explain why AI recommends certain brands more frequently than others.

Something else is happening.

The Familiarity Hypothesis

As we analyzed recommendation patterns, another explanation began to emerge.

Many of the most frequently recommended brands shared one characteristic:

They were already familiar. They had existing brand recognition. Existing awareness. Existing market presence. Existing references across the web.

In other words, recommendation behavior often appeared to resemble memory more than evaluation.

This led us to a framework we call: ## Recommendation by Memory™

Under this model, AI systems frequently recommend brands they have encountered repeatedly during training and exposure.

Not necessarily because those brands are objectively better.

But because they are more familiar.

A Potential Future Shift

However, we do not believe recommendation behavior will remain static.

As AI systems gain access to richer retrieval systems, real-time information, structured commerce data, and increasingly sophisticated evaluation capabilities, recommendation behavior may evolve.

We call this future state: ## Recommendation by Understanding™

Figure 2. The Evolution of AI Recommendation Systems

Recommendation behavior is evolving from what AI remembers to what AI understands. This transition may define the next phase of AI commerce and product discovery.

Under this model, recommendations become less dependent on historical familiarity and more dependent on:

Store quality
Trust signals
Product fit
Verifiable information
Real-time relevance
Semantic understanding

In other words: The best understood businesses may eventually outperform the most familiar businesses.

Why This Matters

If recommendation behavior is driven primarily by memory today, businesses face a difficult challenge.

Optimization alone may not immediately increase recommendation frequency.

But if recommendation systems gradually move toward understanding, the businesses investing in AI readiness today may be building an advantage for tomorrow.

The future of AI commerce may not belong to the brands AI remembers.

It may belong to the brands AI understands best.

The Emerging Question

Most businesses are still asking: Can AI find us?

A more important question may be emerging: Why does AI choose one business instead of another?

That question sits at the center of what we call Recommendation Intelligence™.

And as AI increasingly influences product discovery, recommendation behavior may become one of the most important areas of research in commerce.

About Atom Foundry

Atom Foundry is building the field of AI Commerce Intelligence™.

Our research explores how AI search engines, LLMs, Apple Intelligence, and shopping agents discover, understand, evaluate, trust, recommend, and route customers to businesses.

Current research initiatives include: